How To Create An Opensource NLU API With Rasa
You Have Access To An Exceptional Natural Language Understanding Tool
Introduction
There is much hype regarding chatbots and conversational AI in general.
Technologies are often compared to each other in order to find the best fit for an organization or task.
The problem is that with all the technology available, the initial decisions are easy to make.
The subsequent design & architecture decisions are harder, once the system is established.
In this article I would like to focus on using an opensource option which will serve well for a seed-project.
But more then that, a framework which can evolve and support a fully fledged enterprise solution.
Starting with the NLU API will lay a good foundation to grow the solution into a fully fledged conversational interface.
NLU & Core
There are a few chatbot platforms which have a clear separation between the NLU portion and the dialog management and integration portions. This allows for the development of a stand-alone NLU API.
The Rasa architecture gives you the opportunity to have a NLU API which can also be used for natural language understanding tasks not related to live conversations. This includes conversations archived on email, live agent conversations etc.
NLU Data Format
There are exceptionally good videos by Rasa on how to install in different environments.
Here you can find a video by Rachel for Widows 10 installation.
Secondly, to get started, Rasa has a series of Masterclass videos which is a time efficient way of getting up to speed with the whole Rasa stack.
But, back to creating a NLU API…
The examples below used for NLU training are based on a GitHub project.
The NLU training file has the entities annotated within the intents. Hence we see the healthy trend of intents and entities merging.
Intents and entities are in a text file in markdown format.
Intents & Entities
Simple Examples
The simplest example is just having an intent defined with some examples sans any entities. Below is the check_balance intent with a few examples.
## intent:check_balance
- How much money is on my account?
- what's my balance?
- what's my current balance?
- What's left on that account?
- How much do I have on that account?Let’s add an entity of type account_type with the example credit card.
- what's my [credit](account_type) account balance?The result:
Regular Expressions
This is how regular expressions are defined within the nlu.md file…
## regex:accountNumber
- [0-9]{9}## intent:inform_accountNumber
- my account number is [123454434](accountNumber)
- my account number is [334564343](accountNumber)
- my account number is [940564343](accountNumber)
And the output…
Normalize Data
Data captured in the entity can be normalize. The entity related to to credit cards accounts is captured under the entity of account_type, but always with the value of credit.
{"text":"Pay off my [credit card]{"entity": "account_type", "value": "credit"}, please"}And the result…
Here you see the rather skewed input of creditcard accout is normalized to the value of credit in the account_type entity.
Models
Each time you train your model, a new model file is created in the models folder. This allows you to change between models for your API, roll back to previous versions etc.
Creating The API
With a single command the API is launched on port 5005. But first, make sure you have activated your anaconda virtual environment with:
conda activate rasa2My virtual environment is called rasa2.
Run the API:
rasa run --enable-api -m models/nlu-20200917-225530.tar.gzYou can access the API on the URL
http://localhost:5005/model/parseInteract with your API via a client like Postman.
Conclusion
There are other features also available when defining the training data. The features mentioned here are the primary ones, IMO. In a following article I want to look at the structures available to entities.
